Ethnic variation in validity of classification of overweight and obesity using self-reported weight and height in American women and men: the Third National Health and Nutrition Examination Survey

Obesity, a prevalent and well-established risk factor for cardiovascular disease and diabetes, has been rising in prevalence over recent decades in the United States [1‐7]. In 1999–2000, 64% of U.S. adults were found to be overweight or obese, an absolute increase of 8 percentage points from 1988–1994 [5]. Hence surveillance of obesity prevalence is of great public health interest. Because of the high costs of examination surveys, much of the surveillance of obesity is based on self-reported height and weight data, particularly at the state and local level. A number of studies have reported the validity of self-reported weight and height and BMI computed from them in populations of European extraction in the US and elsewhere [8‐19]. A recent report noted lower validity of self-reported hypertension in Mexican Americans than in non-Hispanics [20]. However, data are lacking on the validity of self-reported obesity data in the growing Hispanic American population.

Therefore a study was done to test the null hypotheses that there is no association of ethnicity with the validity of BMI or "overweight or obesity" determined from self-reported height and weight and that the estimated measure of association of ethnicity with validity of self-reported overweight or obesity is not confounded by age, gender, education, marital status, region, smoking, and health status in the American population. Data on a large, multi-ethnic, national sample of adults from the Third National Health and Nutrition Examination Survey (NHANES III) were examined.

The Third National Health and Nutrition Examination Survey (NHANES III) was conducted in 1988–1994 on a nationwide multi-stage probability sample of 39,695 persons from the civilian, non-institutionalized population aged 2 months and over of the United States excluding reservation lands of American Indians. Mexican Americans and African Americans were oversampled. Of these, 30,818 (78%) were interviewed and examined. Details of the plan, sampling, operation and response have been published as have procedures used to obtain informed consent and to maintain confidentiality of information obtained [20, 21].

Demographic data, years of education completed, medical history including self-assessed health status and behavioral information including smoking history were collected by household interview. Participants chose race and ethnicity categories from a card with categories including Mexican American (MA)[21]. Later in the household interview, interviewers asked, "How tall are you without shoes?" (inches) and "How much do you weigh without clothes or shoes?" (pounds). This occurred before they invited participants to take part in an examination and told them they would be measured and weighed.

Examinations were carried out in a mobile examination center. Technicians measured height to the nearest 0.1 centimeter, weight to the nearest 0.01 kg, as described in detail elsewhere [21‐23]. Body mass index was computed (BMI = weight/height², kg/m²). Obesity was defined as BMI >= 30.0 kg/m². Overweight or obesity was defined as BMI >= 25.0 kg/m² [5]. Extensive descriptive data on height, weight, BMI, and obesity prevalence have been published elsewhere and will not be duplicated here [23‐26].

Table 1 compares prevalence of overweight or obesity (BMI >= 25 kg/m²) based on BMI computed from self-reported height and weight with that based on BMI computed from measured height and weight. In women, prevalence is consistently and substantially underestimated by self-report data, most markedly among MA both in absolute and relative terms. This was also true of MA men. For other men, no consistent difference was seen. Table 2 shows estimated prevalence of obesity (BMI >= 30 kg/m²) by self-report versus measurement. Prevalence is consistently underestimated by self-report, the gap being greater for MA than for other women, but similar for MA and other men. The gap was greater in women than men but varied little with age.

Tables 3 and 4 show test characteristics for the classification of overweight or obesity based on BMI calculated from self-reported height and weight by age, gender and ethnicity. Sensitivity was similar in non-Hispanic European Americans (EA) and non-Hispanic African Americans (AA) but much lower in MA. Sensitivity was also lower in persons aged 60 years and over than < 60 years and in women than men. The highest sensitivity was in younger EA men (93%) and the lowest in older MA women (69%) (Table 3). PPV was similar across ethnic groups. It was higher in older persons and women. PPV was highest in older EA women (98%) and lowest in younger AA and MA men (89%).

Specificity was highest in EA and similar in AA and MA (Table 4). It was higher in women than men and in older than younger men, but varied little with age in women. The highest specificity was in older EA women (98%) and the lowest in younger MA men (82%). In men, PVN was similar in EA and AA but much lower in MA within age groups. It was higher in younger than older men. In women, PVN was highest in EA, intermediate in AA and lowest in MA. It was higher in younger than older women, and lower than in men among AA and MA, but not EA women. The highest PVN was in younger EA women (91%) and the lowest in older MA women (51%).

The effect of smoking status, a possible effect modifier, on sensitivity of overweight or obesity based on self-reported height and weight was examined (Appendix Table A1, see Additional file 1). No consistent effect was seen in men. In women, sensitivity tended to be slightly higher in younger non-smokers than in smokers, but higher in older smokers than non-smokers, the difference being large only among older Mexican American women (smokers 82%, non-smokers 66%).

The percentages not overweight by self report (BMI < 25 kg/m²) with measured BMI >= 25 kg/m², i.e. false negatives for overweight or obesity, among men and women by age, smoking and ethnicity were examined (Appendix Table 2, see Additional file 1). Rates of false negatives by self-report were slightly higher in non-smokers than in smokers, and in MA than in AA or EA. False negative rates were also higher in older than in younger persons and in less educated than more educated persons (not shown).

Tables 5 and 6 show the test characteristics for self-reported data for classifying persons as obese (BMI >= 30 kg/m²). The pattern is similar to that for overweight or obesity. A few exceptions are the similar sensitivity and PVN but lower PPV in MA men compared to other men.

The mean difference between self-reported and measured BMI was greater in Mexican Americans (men -0.37, women -0.76 kg/m²) than in non-Hispanic EA (men -0.22, women -0.62 kg/m²). In a regression model with the difference between self-reported and measured BMI as the dependent variable and MA ethnicity as the exposure variable and controlling for demographic variables or demographic variables plus smoking and health status, MA ethnicity was a significant predictor of BMI difference (Table 7).

An examination of mean differences between measured height and self-reported height indicated that MA men reported themselves to be 0.59 cm, EA men 1.47 cm and AA men 1.16 cm taller than measured. Over-reporting was greater at age 60+y than <60y, e.g. MA 60+ 2.00 cm, 20–59y 0.44 cm. MA men reported themselves to be 0.51 kg, EA men 0.26 kg and AA men 1.09 heavier than measured. Thus underestimates of overweight prevalence in MA men were due to over-reporting of height rather than underreporting of weight.

In women, MA reported themselves to be 0.51 cm, EA 0.26 cm and AA 1.09 cm taller than measured. Over-reporting was greater at age (years) 60+ than <60; e.g. MA 60+ 3.02 cm, 20–59 0.92 cm. MA women reported themselves to be 1.07 kg, EA women 1.21 kg, and AA women 1.58 kg lighter than measured. Under-reporting of weight was greater at age 20–59 than at 60+; e.g. MA 20–59, 1.18 kg; 60+, 0.28 kg. So underestimates of overweight prevalence in each group of women was due primarily to underreporting of weight at younger ages and both to over-reporting height and under-reporting weight at older ages.

Overweight/obesity is one of the leading preventable causes of death in the United States and most industrialized countries [1‐7, 23‐26, 32‐34]. After release of national data from NHANES III (1988–1994) showed striking increases in US obesity prevalence among children and adults compared to earlier surveys, it was widely recognized that an "epidemic" was occurring [23‐26]. Therefore monitoring of the prevalence of obesity by interview surveys, such as the National Health Interview Survey (NHIS) and the Behavioral Risk Factor Surveillance System (BRFSS), which rely on self-reported data, have assumed even greater importance than previously [34]. Yet, all current guidelines for the diagnosis and management of obesity are based on measured height and weight.

Over-sampling of the MA population and availability of measured as well as self-reported height and weight data in NHANES III made it possible to evaluate the ethnic variation in validity of self-reported data. NHANES III data show that using self-reported data produces consistent underestimates of the prevalence of overweight and obesity (BMI >= 25 kg/m²) in MA by up to 21 percentage points and of obesity (BMI >= 30 kg/m²) by up to 14 percentage points (Table 1). Criteria for deciding whether SR data are to be considered valid will depend on the application and on a number of factors including statistical test characteristics and expert judgement as discussed at length elsewhere [35]. However, these data suggest that self-reported data for MA are not sufficiently valid for use in producing such estimates and are suspect for etiologic research. Further research is needed determine the applicability of these findings to data from telephone interviews or self-completed questionnaires.

At a mechanistic level, the important underestimate of BMI and hence overweight and obesity prevalence in NHANES in women was due primarily to under-reporting of weight at younger ages and both to over-reporting height and under-reporting weight at older ages. Possible explanations for under-reporting of weight in women have been discussed [8‐19]. In MA women and men, these may include recall bias and lack of information because of no recent measurements at home or by health care providers. This is important for both weight and height, since stature declines with age in older persons by up to 2 cm per decade after age 30 [8]. Lack of access to scales and health care and/or lack of utilization of either may be due to barriers such as cost, transportation, language or cultural factors. Poorer validity of self-reported hypertension has been noted in MA, perhaps for similar reasons [20]. Further, if obesity and short stature are seen as socially undesirable, a conscious or unconscious tendency to under-report body weight and over-report height may occur even if true weight and height are known. On the other hand, MA women have been reported to be less likely to perceive themselves as overweight and more satisfied with body size than non-Hispanic EA women in some, but not all studies [36‐39]. Specifically in NHANES III, participants were asked, "Do you consider yourself now to be overweight, underweight, or about the right weight?" Hispanic women and men were more likely to under-assess their overweight status than non-Hispanic EA; e.g. 31% of overweight Hispanic women said they were "about the right weight" compared to 14% of non-Hispanic EA [37].

Self reported height, weight and BMI in Mexican Americans may not be of sufficient validity for use by public health agencies and they underestimate obesity prevalence in women of all ethnic groups.